Microstructural porosity segmentation using machine learning techniques in wire-based direct energy deposition of AA6061.

Abstract

Additive manufacturing is promising a flexible and economical alternative to traditional manufacturing. Aviation industry is one such sector that additive manufacturing brought about major transformations. Wire arc additive manufacturing plays an important role in aviation companies to reduce carbon emissions, increase efficiency and allow tailor-made components. However, there are some technical challenges such as microstructural porosity and voids that need to be addressed before they go into full-fledge adoption. Artificial Intelligence (AI) can play a key role in porosity detection. However, applicability of AI for porosity detection is limited due to the difficulties in collecting large amounts of data. The present article demonstrates machine learning models for porosity detection in microstructural images of wire-arc additively manufactured aluminium alloy 6061 parts with limited dataset. Segmentation of pores from microstructures is performed based on pixel-level color and texture features obtained by using Gabor filters. The machine learning models, whose hyperparameters are chosen from cross-validation, achieved an average classification accuracy of 98.89 % (random forest) for porosity detection with pores above the size of 5 μm. Experimental results show that the proposed methods are very effective when compared to the recently proposed methods in the literature.